Loss of regulation of protein synthesis and turnover underpins an attenuated stress response in senescent human mesenchymal stem cells

Significance Cellular senescence and the loss of protein homeostasis are hallmarks of aging, with the former linked to lost efficacy in regenerative medicine and the latter a precursor for age-associated protein folding diseases. Here, using a combination of targeted, -omic, and mathematical approaches, we show how senescent human mesenchymal stem cells are less equipped to resolve the high levels of protein misfolding typical of aged tissue and the post-transplant environment. We report that a lack of translational capacity and a downregulation of protein turnover machinery led to loss of the speed, magnitude, and efficacy of the cellular stress response in senescent cells. Supported by mathematical modeling, these findings challenge a long-standing paradigm of transcriptionally driven regulation of the cellular stress response.


Early passage cells are resistant to inhibition of chaperone activity in the absence of thermal stress.
In order to test the interconnectedness of the chaperone network shown in Figure 2A, we examined regulation of the proteome in cells subjected to targeted inhibition of chaperone machinery. The small molecule 2phenylethynesulfonamide (PES) has been shown to bind selectively to HSPA1A, inhibiting its activity by preventing interactions with its cochaperones (1-4). We found that treatment of EP hMSCs with PES did not greatly perturb the proteome: of 1830 proteins detected with ≥3 peptides-per-protein, only 26 proteins were significantly increased, and 22 decreased (p < 0.05, FDR-corrected ANOVA; SI Fig. S6A). The five major chaperome modules identified in Figure 2A were also not significantly affected (ANOVA, SI Fig. S6B), suggesting that the system was resistant to PES treatment in the absence of additional stress factors.
Nonetheless, we did identify proteins within the HSP70 module that were possibly upregulated as part of a compensatory mechanism, such as heat shock 70 kDa protein 6 (HSPA6) and ER chaperone BiP (HSPA5) (SI Fig. S6C).
To test the generality of this apparent resistance to perturbation, we also examined the effects of targeted inhibition of HSP90 machinery. Gedunin and radicicol both affect HSP90 regulation: gedunin by disrupting interaction with the HSP90-cochaperone prostaglandin E synthase 3 (PTGES3) (5), and radicicol by binding to the N-terminal nucleotide-binding domain of HSP90 proteins, preventing critical ATPase activity (6).

Senescence increases sensitivity to heat stress when combined with inhibition of HSP70.
Having found EP hMSCs to be robust to inhibition of key chaperone proteins in the absence of stress, we applied the same proteomic tools to examine the effects of combining HSP70 inhibition with heat shock and senescence.
Analysis of the effect of PES treatment on EP hMSCs subjected to a 2-hour treatment at 42 °C showed a broader perturbation to the proteome than PES-treated cells maintained at 37 °C (comparing SI Fig. 6K . S6R), and many of the individual proteins within the module appeared to be suppressed (SI Fig. S6S). This is again reminiscent of experiments performed in the absence of PES, where LP cells were unable to mount a chaperome response to thermal stress (Fig. 2F). Interestingly, PES treatments on LP cells at both temperatures caused an upregulation of the CCT/TRiC chaperome module (SI Figs. S6O, R), notably suppressed in senescence (Fig. 2D). This hints at crosstalk between chaperome modules, and a possible role for HSP70 machinery in cytoskeletal maintenance. In addition to demonstrating the application of our network analysis with targeted inhibition of the chaperone machinery by known inhibitors, these results highlight the failure of senescent cells to respond to proteotoxic stress in the same way as early passage cells. Furthermore, this dysregulation was not caused by a desensitization to temperature -the proteome of LP cells was more widely affected by PES treatment at elevated temperature -but rather a failure to remodel specific features of the chaperome.

Primary cell culture.
Human mesenchymal stem cells (hMSCs) were isolated from bone marrow using established methodology (7). The only information associated with each donor sample was source site, donor sex and age; no patient-identifiable information was obtained. hMSCs were cultured on tissue culture treated polystyrene (TCTP) in low-glucose DMEM with pyruvate (Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS, Labtech.com) and 1% penicillin/streptomycin cocktail (PS, Sigma-Aldrich). All quantitative comparisons were made between hMSCs from the same donor, matched at early passage ("EP", proliferating) and late passage ("LP", senescent): a portion of cells were kept in frozen storage whilst donor-   Normalisation was performed as follows: raw peptide ion intensities were log-transformed to ensure a normal distribution and normalised within-sample by equalising sample medians (subtracting sample median). Foldchange differences in the quantity of proteins detected in different samples were calculated by fitting a linear regression model that takes into account donor variability at both the peptide and protein levels (13,14).

Reactome graphs and pathway analysis.
Pathway representation analysis of protein-level fold-changes was carried out using the Reactome Pathway Analysis tool (Pathway browser version 3.6, Reactome database release 74) (15,16). Significantly represented pathways at the FDR-corrected 5% level were coloured according to the mean log2-fold change of proteins within the pathway. Statistical enrichment analysis was carried out using the PANTHER gene list analysis tool (17) to identify over-or under-enriched pathways with an FDR-corrected p-value ≤ 0.05.

Modularity analysis.
The 332 protein human chaperome (18) was modelled as a weighted undirected network in order to discern its community structure. Chaperones with at least one interaction satisfying the highest confidence filter from STRING database (version 11) (19) were used as the nodes of the network, whilst these highest confidence interactions between chaperones were used as undirected edges in the network. Edges between nodes were weighted according to the STRING interaction score between the respective chaperones. The network modularity, ! [−0.5,1], was used to give a measure of how well the network separated into non-overlapping communities of highly-interconnected nodes. A network with edges distributed at random would score a modularity of zero, indicating no presence of community structure, whilst higher scores would indicate more intramodular edges than would be expected at random. The maximal modularity score was calculated using methods described previously (20,21): identified between 181 chaperone proteins. These interactions were used to generate an adjacency matrix , whose elements %& were the weights of edges between two elements , ∈ . To calculate the modularity of the network, was used as the input for the MATLAB Brain Connectivity Toolbox (21). The maximal modularity of the human chaperome was found to be " = 0.5328, with a community structure consisting of 19 modules. The 5 modules with the most proteins seen in our label-free mass spectrometry dataset were investigated and named according to the function of chaperones within the modules.
The response variable -./ (e.g. HSPA1A concentration) was modelled as being dependent on donor d,  (22). It was assumed that there would be no substantial difference between the binding affinity of HSPA1A to HSF1 or client proteins, such that : ≈ ; .
The turnover time for a protein via the proteasome was assumed to be linearly proportional to its length in amino acids. To incorporate this, the ratio between the number of amino acids in HSPA1A and the median length of a human protein (23) was used to estimate the relative time taken for turnover following ubiquitination, < : = = 641: 375 . Parameters were optimised using the MATLAB fminsearch function to minimise the total sum of squared errors (SSE) between in vitro measurements of HSPA1A concentration and in silico values calculated at matched timepoints.
The temporal dynamics of the heat shock response were modelled using MATLAB (R2015a, MathWorks). All populations were recalculated at discrete time intervals using ordinary differential equations (ODE), except for 70 which was calculated using a delay differential equation (DDE). This accounted for the time delay ( ~ [60, 180] minutes) between transcription and translation (24). The ODE/DDEs were evaluated every (simulated) 0.01 minutes. A proteotoxic stress comparable with heat shock was simulated by multiplying the reaction rate > by another optimised parameter, > 1 for 120 minutes.

Protein labeling with monobromobimane (mBBr). Media was removed from cells in T75 flasks (Corning)
either pre-stress or immediately post-stress and cells were washed with PBS. Cells were then labelled by incubation with 5 mL of 400 µM monobromobimane (mBBr; Sigma-Aldrich) in PBS at 37 °C for 10 min.
Following labelling, 5 mL of 2 mM glutathione in PBS was added to quench the reaction. The quenched mBBr solution was removed and cells washed with PBS, before harvesting with trypsin for MS analysis, as described above. When searching MS data, the peptide database was modified to search for mBBr-adducts to cysteine (monoisotopic mass changes, 133.053 and 150.056 Da). mBBr labelled peptides were filtered to only include sequences from reviewed protein annotations (25) and labelled peptides with missed cleavages were summed with their fully tryptic counterparts. 359 unique mBBr labelled peptides were detected. The log2 fold-change in labelling across samples was normalised to the respective log2 fold-change in protein abundance. A Wilcoxon signed-rank test was used to determine whether the mBBr-labelling profile changed between samples at the 95% confidence level.

RNA-sequencing.
All reagents and solutions were supplied with the TruSeq Stranded mRNA assay (Illumina, Inc.,#20020594). Quality and integrity of total RNA samples from five biological replicates were assessed using a TapeStation 2200 (Agilent Technologies). To generate libraries, total RNA (0.1 -4.0 µg) was used as input material from which polyadenylated mRNA was purified using poly-T, oligo-attached, magnetic beads.
The mRNA was then fragmented using divalent cations under elevated temperature and then reverse transcribed into first strand cDNA using random primers. Second strand cDNA was then synthesised using DNA Polymerase I and RNase H. Following a single 'A' base addition, adapters were ligated to the cDNA fragments, and the products then purified and enriched by PCR to create the final cDNA library. Adapter indices were used to multiplex libraries, which were pooled prior to cluster generation using a cBot instrument i.e. pre-stress). (G) Expanded view of "metabolism" and "transport of small molecule" pathways, following Reactome pathway analysis of the proteins shown in Figure 1D of the main paper, comparing the responses of EP and LP hMSCs subjected to heat shock (16). Significantly represented pathways (FDR-corrected pvalue < 0.05) are shown in colours corresponding to the mean log2 fold-change of proteins in the pathway.
The sub-pathways "metabolism of carbohydrates" and "iron uptake and transport" were significantly overenriched in LP hMSCs, but there were no overall significant changes to the parent pathways.  (14); n = 4 primary donors. Significance of changes to modules was determined from ANOVA testing. In network diagrams, node size is indicative of chaperone degree, while edge weight indicates interaction score between chaperones. The central node is coloured according to the mean abundance change of chaperones associated with the module. Descriptions of the chaperone modules can be found in Fig. 2A of the main paper.